Impact Evaluation of Score Classes and Annotation Regions in Deep Learning-Based Dairy Cow Body Condition Prediction

SIMPLE SUMMARY: The body condition of dairy cattle is an essential indicator of the energy supply of the animals. Various scoring systems are used in practice to quantify body condition. These systems rely on visual observation of different body parts and sometimes on collecting tactile data. In all cases, scoring requires expert knowledge and practice and is time-consuming. Therefore, it is rarely carried out on livestock farms. However, for animal husbandry and veterinary practice, it would be meaningful to have data on the condition of the animals continuously or even daily, which is not feasible with expert scoring. We investigated how computer vision-based supervised deep learning, specifically neural networks, can automate body condition scoring. To execute this, we have used video recordings of the rumps of a large number of animals. We have trained and tested various convolutional neural networks with this collected data. Scoring by trained networks yielded results that met or exceeded the agreement among experts. We have made our trained neural networks freely available, using these as pretrained models. Those working on similar developments can achieve even better results with less data collection required with their own fine-tuning. ABSTRACT: Body condition scoring is a simple method to estimate the energy supply of dairy cattle. Our study aims to investigate the accuracy with which supervised machine learning, specifically a deep convolutional neural network (CNN), can be used to retrieve body condition score (BCS) classes estimated by an expert. We recorded images of animals’ rumps in three large-scale farms using a simple action camera. The images were annotated with classes and three different-sized bounding boxes by an expert. A CNN pretrained model was fine-tuned on 12 and 3 BCS classes. Training in 12 classes with a 0 error range, the Cohen’s kappa value yielded minimal agreement between the model predictions and ground truth. Allowing an error range of 0.25, we obtained minimum or weak agreement. With an error range of 0.5, we had strong or almost perfect agreement. The kappa values for the approach trained on three classes show that we can classify all animals into BCS categories with at least moderate agreement. Furthermore, CNNs trained on 3 BCS classes showed a remarkably higher proportion of strong agreement than those trained in 12 classes. The prediction precision when training with various annotation region sizes showed no meaningful differences. The weights of our trained CNNs are freely available, supporting similar works.